The monthly natural gas balancing item, which reflects the difference between consumption and supply at the national level,1 is a significant metric for gauging the comprehensiveness and quality of EIA’s volumetric data as presented in the Natural Gas Monthly (NGM). A negative balancing item is the result of supply that is greater than consumption. A positive balancing item is indicative of consumption that exceeds supply. Historically, the balancing item has exhibited a consistent seasonal pattern, with considerable spikes evident at certain times during the year. In the fall, reported supply exceeds reported demand and the balancing item trends significantly negative. In spring the opposite occurs, as reported demand exceeds reported supply, resulting in a substantial positive balancing item. This phenomenon is depicted in Figure 1. Other factors contribute to the monthly imbalance, but the seasonal aspect warrants particular attention because it has had a significant and systematic effect over the years.
The discernible pattern in Figure 1 suggests a flaw in the data collection process behind the NGM. Research conducted via site visits to several respondents that file Form EIA-857, "Monthly Report of Natural Gas Purchases and Deliveries,” EIA’s primary monthly natural gas consumption survey, indicates that many local distribution companies (LDCs)2 are reporting end-use deliveries to residential and commercial customers on a lag. That is, they are providing data corresponding to their billing cycles rather than to the actual calendar month being sought by EIA. This in turn creates a disconnect between supply and demand patterns at the published aggregate level, as estimated supply precedes estimated demand going into and out of the heating season each year.
The lag in consumption data is a direct result of LDC business practices. Typically, LDCs read a separate grouping of residential and commercial customer meters on each business day during a calendar month. In order to submit Form EIA-857, a number of respondent LDCs sum the usage data from the meters read during the calendar month and submit this total to EIA. In cases where meters are read near the end of the report month, usage somewhat aligns with the actual calendar month. However, meters read early in the month contain data more representative of the prior month’s usage. On average, about half the billing volume is used during the report month with the other half used in the month preceding the report month.
Because residential and commercial customers utilize natural gas primarily for space heating, weather patterns drive demand in these sectors. It is during the spring and fall transitional periods that supply and demand become incrementally disjointed in the NGM, due in large part to the effects of lag reporting. As noted above, this phenomenon primarily affects the residential and commercial sectors.
The relationship between the balancing item and month-to-month heating degree-day3 differential – or change in HDD over the course of a month - serves to illustrate the seasonal impact of lag reporting (Figure 2). As shown in the figure, when heating degree-day differential is positive, the prior month was warmer than the report month, suggesting under-reporting of consumption and in turn a negative balancing item. Negative heating degree-day differential correlates to a colder prior month, suggesting over-reported consumption and a resultant positive balancing item. Furthermore, a regression analysis performed on the same dataset shows that the difference in heating degree-days accounts for nearly 80 percent of the total variability in the balancing item.4
It is worth noting that variation in the balancing item is not universally attributable to lag reporting. The NGM relies on additional sources apart from the EIA-857 to cover the remaining segments of the natural gas industry, including production, imports and exports, underground storage, and electric power consumption.5 These data are currently being reviewed as part of an ongoing effort to identify and rectify other causes of balancing item behavior outside of the seasonal phenomenon.6 Nevertheless, improvements in calendar month consumption data would provide a more accurate, synchronous snapshot of gas markets in the United States.
Addressing the Problem
EIA staff made site visits to a set of EIA-857 respondents who together accounted for nearly a third of total U.S. residential and commercial deliveries of natural gas. These visits provided important insights into industry recordkeeping practices and available data resources. In discussing reporting methodologies, for example, it was learned that along with the billing cycle data that form the basis for many EIA-857 submissions, total system sendout7 is another widely available element. This number, which is generally available on a calendar month basis, captures natural gas going into the distribution system without specifying the end-use sector for which it is intended. By aligning the various available data sources in Figure 3, a representation of LDC reporting practices and methodologies can be visualized.
The graphic on the left illustrates lag effect inherent in companies reporting calendar month usage from meters read during the month. A billing cycle customer grouping with meters read on the first of the month would consist of data largely from the month prior to the report month. In the diagram this is represented by a horizontal slice from the bottom layer of the parallelogram. Similarly, a group with meters read on the final day of the month would represent a horizontal slice from the top layer of the figure, with most of the usage falling in the report month. Alternatively, the graphic on the right, sendout data would provide a more accurate measure of calendar month use; however, the end-use sector of the natural gas can not be precisely determined by this measure alone.
While discussions with select LDCs revealed that not all are reporting billing cycle (i.e., lagged) data on the EIA-857, evidence suggests that the proportion who do is significant. The total percentage of residential and commercial volume reported in this manner is perhaps on the order of 70 percent. Over the course of these site visits, two other general reporting methods were indicated apart from straight billing cycle data: use of an estimation technique, e.g., sample meters or heating degree-day-based models, or the allocation of the sendout volume proportional to the billing cycle usage. This latter method forms the basis of a proposed solution going forward.
EIA’s goal in identifying an alternate data collection and estimation protocol is to more precisely align monthly consumption estimates with the other aspects of natural gas supply and disposition such that all data correspond to a calendar month timeframe. From discussions with several LDCs, it became apparent that the total sendout volume was the most expedient means of achieving a meaningful improvement in the published estimates for two major reasons:
In March 2010, EIA was granted Office of Management and Budget (OMB) approval to add a line item for total sendout to Form EIA-857 (Figure 4). This change, along with refinement of the instructions to clearly instruct that unadjusted billing cycle data be reported on lines 3 through 12 will provide the basis for better consumption estimates using an approach, detailed here, that consistently accouts for the lag between sendout and billing data for all respondents.
- It represents an accurate calendar month measurement of total end-use deliveries.
- The number is already available from other business processes. Therefore collection results in a minimal increase in reporting burden.
At the respondent level, sendout data will become the new basis for total consumption. The majority of the natural gas in the electric power and industrial sectors is transported by LDCs and pipelines on behalf of third party suppliers. Meters for these customers are generally read daily or at the end of the calendar month so that usage corresponds to the actual calendar month, and in turn, the third party supplier’s nomination cycle. Thus, the timeliness of these sectors allows their totals to be subtracted from the total sendout, with the remaining volume, (minus a small amount estimated for distribution use and/or loss), attributable to the residential and commercial sectors. The residential and commercial sector shares of the combined residential and commercial sendout will be determined by the proportion of use from the most recent billing data reported on lines 3 through 12 on the form. For example, if the residential sector accounted for 60 percent of the total residential and commercial volume, the residential sector would receive 60 percent of the remaining sendout volume. The end result of this adjustment is that the sector totals sum to the sendout volume, which reflects actual monthly consumption.
As Figure 3 depicts, the use of sendout volumes will enable EIA’s consumption data to correspond more closely to the calendar month and, in turn, EIA’s other supply and disposition series. This will not only mitigate the seasonal aspect of the balancing item, it will also create a uniform data reporting scheme across all EIA-857 survey respondents.
The EIA-857 reported electric power totals are used in the process of allocating sendout to the residential, commercial, and industrial sectors; however, EIA’s official electric power consumption data are from a source other than the EIA-857.8 In line with the current practice, the State-level electric power volume from the official source will be combined with the EIA-857 based residential, commercial, and industrial volumes to create State-level consumption totals. A final note of interest on the new procedure is that sector-level prices as published in the NGM would still correspond to the LDC billing cycles, as billing data represent the only source of an LDC’s revenue stream.9
Effectiveness of Adjustment: National Level Example
An example serves to illustrate the impact of the proposed methodology change. While estimating the precise effect at the national level is difficult, it is possible to follow a similar process using citygate volume data from the EIA-857 to extrapolate the hypothetical result (EIA currently collects citygate volumes for the purpose of calculating prices). For this analysis, criteria were set to filter out data that did not meet certain quality standards.10 The estimated volumes based on using citygate as a proxy for total sendout can be compared with volumes as currently reported (Figure 5). As expected, estimated consumption is lower than reported consumption during the spring, when the reported value contains data from the previous month that was likely cooler than the report month. Similarly, estimated consumption is higher than reported consumption during the fall, when the reported total contains data from the previous month that was likely warmer than the report month. The green bars at the bottom of the figure represent the difference between the estimated and reported volumes.
The differences displayed in Figure 5 can be grossed up to provide a hypothetical change in monthly consumption at the national level, which is then applied to the balancing item (Figure 6). It is important to note that there are several assumptions11 used to construct this hypothetical balancing item, and the results are intended merely to provide an approximation of the expected changes. Further, while citygate data are useful as a proxy in this analysis, they are not synonymous with total sendout, as citygate includes only natural gas that is owned by the LDCs and does not include natural gas for transportation to end users, which constitutes a significant share of total consumption.
In Figure 6, the navy lines represent the current balancing item, while the red lines represent the estimated balancing item after the effects of billing cycle data are accounted for. Although the results do not appear random, the seasonality has been largely removed. Randomness would not be the expected end result, as there are other places in the data where estimation procedures may lead to errors that would be expected to persist across months. Nonetheless, the average reduction of the balancing item in absolute terms is around 35 billion cubic feet per month.
EIA has developed an implementation timeline for the new data collection and estimation methodology. Under this plan, respondents to Form EIA-857 will receive official notification of the change along with updated forms and instructions by mid-summer 2010. The first month for which sendout data will be collected is August 2010. These forms will be due to EIA by September 30, 2010, meaning that the first published aggregates featuring the new methodology would appear in the October NGM. Thus the new methodology will be in place by fall of 2010, more closely aligning EIA’s supply and consumption data heading into the next heating season.
While EIA recognizes that the proposed methodology will not fully eliminate the balancing item, it represents a practical measure to improve published data without substantially increasing the reporting burden of survey respondents. It is believed that the transition will yield a significant benefit to EIA’s user community.